Est.

Completed Operations Coverage Requirements in Robotics Deployment Agreements

Robotics vendors need more than a certificate to survive enterprise insurance exhibits.

· 9 min read
Cover illustration for “Completed Operations Coverage Requirements in Robotics Deployment Agreements”
MSA Insurance Requirements · November 9, 2026 · 9 min read · 1,920 words

Completed operations coverage requirements in robotics deployment agreements demand more than a general liability certificate. For hardware and robotics startups reviewing an enterprise agreement, the decisive questions are whether additional insured protection extends to completed operations, whether AI exclusions undermine that protection, and whether policy terms address autonomous behavior after delivery. Those questions must be resolved before the insurance exhibit is finalized.

Why enterprise deployment agreements now contain completed operations clauses that most robotics vendors have never encountered before

Enterprise buyers signing up for multi-year robotics deployments want insurance exhibits that cover the risks of running the robots as well as allowing them onto a site. Their agreements import language from construction, defense, and heavy industry, where a finished job can cause injury or property damage years after the contractor leaves. A certificate alone does not satisfy that concern. Buyers want the contract to protect them against losses that arise after installation, commissioning, and acceptance, when the vendor might otherwise see its deployment work as finished.

That requirement changes how a robotics company used to pilot-stage agreements negotiates the deal. The insurance exhibit is not merely an administrative attachment to agreed pricing and delivery terms. The insurance exhibit sets the risk protection requirements the vendor must meet for the customer. A deployment can therefore have agreed business terms and still be held up because the insurance program does not meet the exhibit’s requirements.

The difficulty is more than unfamiliar terminology. Unlike a building contractor, a robotics vendor leaves behind a machine that keeps running and whose behavior it may still shape. The robot continues to generate liability exposure while the vendor supplies updates or the customer changes operating parameters. Insurance language written for a finished job must therefore cover a system that can change how it behaves after handover.

What "completed operations" actually means and where it comes from

Completed operations coverage addresses bodily injury or property damage that occurs after a contractor has finished its work and handed the job site back. It covers risks that ongoing operations protection may leave out. The distinction is temporal, but it also concerns the status of the work. An incident during installation and an incident caused by the finished installation are not the same insurance problem.

The classic example is a subcontractor installing a component. The job is completed and signed off, the subcontractor leaves, and the component fails months or years later, injuring someone. The work is no longer an ongoing operation. Completed operations coverage is designed to respond to that later harm, subject to the policy’s terms and exclusions.

The owner wants protection under the vendor’s policy if the vendor’s completed work later causes harm, including defense and indemnification where the coverage applies, avoiding the need to pursue the vendor separately. For construction, a finished installation provides a recognizable boundary. For robotics, hardware acceptance may mark a stage in the deal while leaving open when the vendor’s work on the system ends, so the comparison with construction only goes so far.

Why the "completed" trigger is genuinely ambiguous when the robot keeps running after deployment

A robot can be delivered and commissioned while remaining subject to remote updates, retraining on new data, and customer reconfiguration. Completion is therefore not a single obvious engineering event. A later injury may arise from the original deployment, subsequent work, or changes made by another party.

Consider a robot that causes harm six months after deployment following an update pushed remotely by the OEM. Its physical installation is complete, but the behavior that led to the loss may stem from work done later. Detailed logs are needed to tell which work led to the loss. Without them, the boundary between completed and ongoing work becomes a dispute about the system’s history, with the date on an acceptance document insufficient to settle it.

If the site operator modifies programming, operating parameters, or safety systems, liability may shift, but a clause naming the OEM’s policy does not decide who bears the loss. The standard industry position combines CG 20 10 and CG 20 37 for ongoing and completed operations additional insured protection. Yet there is currently no ISO AI endorsement providing completed operations coverage to a party with no privity and no work being performed for them. That gap cannot be repaired through contract wording alone.

How AI exclusions in standard GL policies compound the completed operations problem

Even where the completed operations endorsement is present, an AI exclusion may undermine the protection it appears to provide. Broad generative AI exclusions can bar coverage when AI contributes to a loss, even when it is one of several causes. The endorsement and exclusion must be read together. Establishing that an incident falls within completed operations does not establish that the policy covers its cause.

For a robotics deployment, that distinction is central. A claim involving an AI decision, a software failure, or behavior traceable to autonomy may raise the very exclusion that the insurance exhibit never addresses. Whether an exclusion applies depends on what it says and what happened. The presence of software or autonomy is not a substitute for reading that language, but neither is a completed operations endorsement proof of affirmative protection.

This creates a mismatch between contractual compliance on paper and protection against a loss. The vendor may provide evidence of general liability insurance and the requested additional insured form while leaving both itself and the site owner exposed to an excluded claim. Form names alone cannot resolve that mismatch. The policy’s treatment of autonomous operations must be examined with its completed operations provisions, rather than treated as a separate technology insurance question.

What insurance clauses should a robotics startup watch out for in its first enterprise or government MSA?

General liability protects against covered liability claims, with the site owner named as an additional insured for both ongoing and completed operations. Workers’ compensation addresses staff working on site. Equipment coverage is for machines left at customer facilities, while umbrella coverage comes into play when the required liability limits are higher than the primary policy’s limits.

The additional insured language deserves particular attention. A requirement for CG 20 37 calls for completed operations protection that goes beyond the ongoing operations protection in CG 20 10. The contract’s allocation of responsibility and the policy’s promise of coverage are separate, so an obligation to indemnify does not itself establish that insurance will fund the loss.

Government contracts can add further layers. Some call for tech E&O by name, while defense-adjacent agreements may call for cyber coverage that covers incidents involving IoT or robotic systems, which generic cyber coverage that excludes connected hardware would leave out. Those requirements cannot be answered by a general liability certificate.

Higher limits also require a substantive review. An enterprise agreement may demand total liability limits several multiples above the startup’s primary coverage, but umbrella policies have their own AI exclusion exposure. An umbrella that follows the underlying exclusion does not restore protection for the excluded claim.

Why the underwriting submission determines whether completed operations coverage actually holds for autonomous systems

The underwriting submission gives the insurer the facts it needs to judge the risks of autonomous operations before the policy binds. Standard general liability submissions ask generic questions about operations, revenue, and premises. They do not adequately describe autonomy levels, remote update protocols, multi-party liability chains, or AI governance documentation.

A submission prepared for a carrier with explicit appetite for autonomous operations addresses those risks directly. It connects the description of the business to how the deployed system operates and changes after delivery. That detail supports a policy designed for the actual exposure. It also lets the completed operations requirement and any AI exclusion be reviewed together before coverage is bound, though the wording in the issued policy still decides what is covered.

System observability is part of this underwriting problem. Specialist carriers require immutable logs of autonomous actions for policy renewal because forensic investigators need to distinguish a prompt injection attack, a training data flaw, and an environmental anomaly. Those records also support the causal analysis a completed operations claim requires. The same evidence that helps explain a failure can help establish whether it arose from original deployment work or a later intervention.

The submission therefore does more than describe the company. It makes the changing system legible to the insurer, reducing the risk that autonomous operations are placed into coverage designed for a conventional business.

The timing problem — when to secure completed operations coverage relative to signing a deployment agreement

Many robotics vendors encounter the insurance exhibit only after commercial terms are agreed and the pilot is weeks away. At that point, an unfamiliar endorsement becomes a scheduling problem. Placements involving autonomy take longer than standard placements because fewer carriers have appetite for the risk. There may be too little time left to describe the operations properly, deal with exclusions, and get the required coverage.

A missing certificate can then delay the pilot, but the certificate is only the visible symptom. The underlying problem may be a completed operations endorsement the existing broker cannot produce or a policy exclusion that makes the proposed protection unsuitable. Issuing paperwork cannot fix missing coverage. Nor does a signed deployment agreement make its insurance requirements available in the market.

The appropriate sequence is to engage a specialist broker before the insurance exhibit is finalized. At that stage, the broker can identify non-standard requirements, help negotiate wording the carrier can satisfy, and examine whether an AI exclusion offsets the completed operations endorsement.

This is a technical sequencing requirement rather than a general argument for more brokerage services. The contract defines the protection the buyer expects, while underwriting establishes what the insurer is prepared to cover. Resolving the difference before signature avoids making deployment contingent on insurance terms that have not yet been secured.

What robotics vendors should verify in their current policies before the next contract arrives

The immediate task is to read the full policy and its endorsements before the next enterprise negotiation. The declarations page and certificate cannot answer these questions. The review should establish whether CG 20 37 provides completed operations additional insured protection, rather than relying only on CG 20 10 for ongoing operations. It should also compare available limits with the requirements the enterprise agreement will impose.

The exclusion review should identify CG 40 47, CG 40 48, CG 35 08, and any manuscript AI exclusion language. Where an exclusion is present, the question is whether it has been affirmatively carved back for the vendor’s specific operations. The wording must address the actual deployment. A generic description of the business cannot resolve an exclusion aimed at the behavior that may cause the claim.

The definitions of “your work” and “your product” also require examination. They must be read with autonomous software behavior after hardware delivery in mind, since the installed system can keep changing. The umbrella needs its own review. If it follows the underlying AI exclusion, it provides no incremental protection for the excluded AI-related completed operations claim, regardless of the additional limits it offers.

Finally, where tech E&O is part of the program, the review should establish whether it explicitly covers physical harm caused by autonomous behavior in addition to economic loss from software errors. These are questions for a licensed specialist broker before binding, supported by a submission that explains the system accurately. A certificate may open the customer’s gate. Protection for what the robot does after it enters is subject to the full policy’s terms.

Sources

  1. The Insurance Industry’s Retreat from AI Threatens to Slow Innovation and Adoption

More in MSA Insurance Requirements